Immunogenicity prediction method for antibody drug development
Through machine learning ensemble model based on the amino acid sequence of the antibody variable region, the immunogenicity of antibody drugs can be quickly and accurately predicted, and the problems of low accuracy and high cost of prediction methods in the prior art are solved, and the screening efficiency of drug development is improved .
Patent Information
- Application Number
- PCT/CN2024/092561
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2023-05-11
- Filing Date
- 2024-05-11
- Publication Date
- 2024-11-14
AI Technical Summary
In the development of existing antibody drugs, there is a lack of fast and accurate immunogenic prediction methods, which leads to great risks in drug design and development. Traditional methods require three-dimensional structure input, which is costly and long cycles.
Based on the amino acid sequence of the antibody variable region, the features were extracted using the ProtT5-XL-UniRef50 pre-trained model, combined with Sentence-T5 Encoder-only means and PCA dimensionality reduction, and the machine learning integration model was constructed using AutoGluon to integrate multiple sub-models to perform antibody immunogenicity. predict.
Fast and accurate antibody immunogenicity prediction is achieved, which reduces time, material and labor costs, improves the screening efficiency of drug development, and is significantly better than the prediction accuracy of existing methods.
Smart Images

Figure CN2024092561_14112024_PF_FP_ABST
Abstract
Description
Immunogenicity prediction methods for antibody drug development
[0001] CROSS-REFERENCE TO RELATED APPLICATIONS
[0002] This application claims priority to Chinese patent application No. 202310530327.9 filed on May 11, 2023, the entire contents of which are incorporated herein by reference. Technical Field
[0003] The present disclosure relates to antibody drug development, and more specifically to an immunogenicity prediction method based on an integrated model of antibody variable region amino acid sequence and machine learning for antibody drug development. Background Art
[0004] Therapeutic proteins include monoclonal antibodies, coagulation factors, hormones, growth factors and plasma proteins, and are a rapidly growing segment of the pharmaceutical industry. In recent years, monoclonal antibodies have contributed nearly half of the therapeutic proteins approved by the U.S. Food and Drug Administration (FDA). Approved therapeutic proteins are applicable to a variety of fields, such as cancer, autoimmune diseases, inflammation and genetic diseases. However, when therapeutic proteins are used to treat patients, unnecessary immune responses, such as the production of anti-drug antibodies (ADA), may cause a variety of problems, including changes in pharmacokinetics, loss of efficacy, and even life-threatening complications. Therefore, low immunogenicity is an important issue that needs to be considered in the drug development process of therapeutic antibodies. The development of computational tools to predict antibody immunogenicity based on antibody sequences is of great significance for screening in the early stages of drug design and reducing the risk of therapeutic antibody development.
[0005] Most of the existing computational tools for antibody immunogenicity predict B cell epitopes and T cell epitopes, and are unable to obtain the clinical immune response of antibody drugs after injection into the body from a macroscopic perspective. There are very few tools for predicting the clinical immunogenicity of antibody drugs. Shide Liang et al., based on the characteristics of B cell epitopes, including the cavity volume and hydrophobicity of the CDR-H3 loop and the number of glycine at the CDR-H2 corner, used a support vector machine (SVM) to construct a classification model using the leave-one-out cross-validation method and tested it on an independent test set to develop a prediction tool PITHA for distinguishing high / low antibody immunogenicity. For details, see Shide Liang, Chi Zhang, “Prediction of immunogenicity for humanized and full human therapeutic antibodies”, PLoS One, 2020, 15(8): e0238150; doi: 10.1371 / journal.pone.0238150. For more information on this prediction tool, see International Patent Publication WO2021 / 147642A1. The entire contents of the above-mentioned documents and patent publications are hereby incorporated into the present disclosure by reference, and become part of the contents of the present disclosure.
[0006] Although the above methods have been developed to predict the high / low clinical immunogenicity of therapeutic antibody drugs, they require the three-dimensional (3D) structure of the antibody as input to extract structural features. However, the acquisition of 3D structure is not easy. Traditional experimental methods to obtain 3D structure are expensive and time-consuming, and computational structure prediction methods still have certain errors. In addition, the data sets used to construct the models of existing methods are small, and the accuracy of the predictions also has a lot of room for improvement. Therefore, it is necessary to further develop faster and more accurate computational methods based on antibody sequence to predict its immunogenicity.
[0007] Summary of the Invention
[0008] In response to the limited data usage, low accuracy, and computational time required for predicting the immunogenicity of therapeutic antibodies currently available, this disclosure proposes a method for predicting immunogenicity for antibody drug development. Simply by providing the amino acid sequence of the antibody's variable region, this method can quickly and accurately predict the high or low immunogenicity of the antibody. This aids immunogenicity screening in the early stages of therapeutic antibody drug development, significantly saving time, materials, and labor costs.
[0009] According to a first aspect of the present disclosure, a method for predicting antibody immunogenicity is provided. The method may include: extracting features based on the amino acid sequence of the antibody variable region; and inputting the extracted features into an antibody immunogenicity prediction model based on a machine learning ensemble model to obtain a prediction result of the antibody immunogenicity.
[0010] In the method according to the first aspect of the present disclosure, preferably, the feature extraction based on the antibody variable region amino acid sequence may include: extracting a feature vector for the antibody heavy chain variable region amino acid sequence; extracting a feature vector for the antibody light chain variable region amino acid sequence; and splicing the feature vector extracted for the antibody heavy chain variable region amino acid sequence and the feature vector extracted for the antibody light chain variable region amino acid sequence to obtain the feature vector of the antibody.
[0011] Preferably, extracting a feature vector for an antibody heavy chain variable region amino acid sequence may include: using the antibody heavy chain variable region amino acid sequence as input, extracting a feature matrix using a protein pre-trained model; and compressing the extracted feature matrix into a feature vector. Similarly, extracting a feature vector for an antibody light chain variable region amino acid sequence may include: using the antibody light chain variable region amino acid sequence as input, extracting a feature matrix using a protein pre-trained model; and compressing the extracted feature matrix into a feature vector.
[0012] Preferably, the protein pre-training model can be implemented by a combination of one or more of the following models: ESM-1b, UniRep, ProteinBert, TAPE, ProtGPT2, ProtTXL, ProtBert, ProtXLNet, ProtAlbert, ProtElectra, ProtT5-XL-BFD, ProtT5-XL-UniRef50 and ProtT5-XXL; more preferably, the protein pre-training model is ProtT5-XL-BFD or ProtT5-XL-UniRef50; more preferably, the protein pre-training model is ProtT5-XL-UniRef50.
[0013] Further preferably, the feature vector extraction for the antibody heavy chain variable region amino acid sequence may include: taking the antibody heavy chain variable region amino acid sequence as input, and using the ProtT5-XL-UniRef50 pre-trained model to extract the embedding layer vector output by the encoder, wherein each amino acid corresponds to a 1024-dimensional vector, and each sequence corresponds to an L×1024-dimensional feature matrix, where L is the length of the antibody heavy chain variable region amino acid sequence; averaging the above feature matrix by column to obtain a 1024-dimensional feature vector. Similarly, the feature vector extraction for the antibody light chain variable region amino acid sequence may include: taking the antibody light chain variable region amino acid sequence as input, and using the ProtT5-XL-UniRef50 pre-trained model to extract the embedding layer vector output by the encoder, wherein each amino acid corresponds to a 1024-dimensional vector, and each sequence corresponds to an L'×1024-dimensional feature matrix, where L' is the length of the antibody light chain variable region amino acid sequence; averaging the above feature matrix by column to obtain a 1024-dimensional feature vector.
[0014] Preferably, the Sentence-T5 Encoder-only mean method is used to average the values of each dimension of the embedding layer feature values of all amino acids output by the encoder by column to obtain a 1024-dimensional feature vector.
[0015] In the method according to the first aspect of the present disclosure, preferably, the 1024-dimensional feature vector obtained by extracting the amino acid sequence of the antibody heavy chain variable region and the 1024-dimensional feature vector obtained by extracting the amino acid sequence of the antibody light chain variable region can be spliced to obtain a 2048-dimensional feature vector.
[0016] In the method according to the first aspect of the present disclosure, preferably, a principal component analysis (PCA) model can be used to perform dimensionality reduction processing on the feature vector of the antibody to obtain the feature vector of the antibody after dimensionality reduction.
[0017] In the method according to the first aspect of the present disclosure, preferably, the antibody immunogenicity prediction model based on the machine learning integrated model can be constructed by the following steps: obtaining a data set, wherein the data set includes multiple training samples, each training sample includes the variable region amino acid sequence and immunogenicity data of the sample antibody; performing feature extraction on the variable region amino acid sequence of each of the sample antibodies; using the extracted features as input data and the antibody immunogenicity data as output data of the machine learning integrated model, training the machine learning integrated model, and obtaining a final antibody immunogenicity prediction model based on the machine learning integrated model.
[0018] Preferably, the training of the machine learning ensemble model may include: using an automated machine learning method to construct and train the machine learning ensemble model based on one or more machine learning models. In some embodiments, the machine learning model may be a deep learning model. In some embodiments, the automated machine learning method may be selected from AutoGluon (https: / / auto.gluon.ai / stable / index.html), Auto-Sklearn (https: / / automl.github.io / auto-sklearn / master / ), TPOT (http: / / epistasislab.github.io / tpot / ), Hyperopt Sklearn (https: / / github.com / hyperopt / hyperopt-sklearn), Auto-Keras (https: / / autokeras.com / ), H2O AutoML (https: / / docs.h2o.ai / h2o / latest-stable / h2o-docs / automl.html), and TransmogrifAI (https: / / transmogrif.ai / ). In some specific embodiments, the automated learning method may be AutoGluon.
[0019] Preferably, the AutoGluon tool can be used to build and train the machine learning integration model based on one or more models of NeuralNetFastAI, LightGBMLarge, CatBoost, LightGBM, XGBoost, LightGBMXT, NeuralNetTorch, KNeighborsUnif, KNeighborsDist, ExtraTreesEntr, ExtraTreesGini, RandomForestGini and RandomForestEntr models.
[0020] More preferably, the AutoGluon tool can be used to build and train the machine learning integration model based on the XGBoost and NeuralNetFastAI models.
[0021] According to a second aspect of the present disclosure, a method for constructing an antibody immunogenicity prediction model based on a machine learning ensemble model is provided. The model construction method may include: obtaining a data set, the data set including multiple training samples, each of the training samples including a variable region amino acid sequence and immunogenicity data of a sample antibody; extracting features from the variable region amino acid sequence of each sample antibody; using the extracted features as input data and the antibody immunogenicity data as output data of the machine learning ensemble model, training the machine learning ensemble model, and obtaining a final antibody immunogenicity prediction model based on the machine learning ensemble model.
[0022] According to a third aspect of the present disclosure, a system for predicting antibody immunogenicity is provided. The system may include: an acquisition module for acquiring a sample to be predicted, wherein the sample to be predicted includes the variable region amino acid sequence of the sample antibody to be predicted; a processing module for obtaining a prediction result of the immunogenicity of the sample antibody to be predicted by inputting the variable region amino acid sequence of the sample antibody to be predicted, wherein the processing module further includes the following submodules: a feature extraction submodule for extracting features based on the variable region amino acid sequence of the sample antibody to be predicted; and a model prediction submodule for inputting the extracted features into an antibody immunogenicity prediction model based on a machine learning integrated model to obtain a prediction result of the antibody immunogenicity.
[0023] The system for predicting antibody immunogenicity according to the third aspect of the present disclosure can be implemented by a computer. Preferably, the following operations can be implemented by executing a computer program: obtaining a sample to be predicted, wherein the sample to be predicted includes the variable region amino acid sequence of the sample antibody to be predicted; obtaining a prediction result of the immunogenicity of the sample antibody to be predicted by inputting the variable region amino acid sequence of the sample antibody to be predicted, and the operation further includes the following sub-operations: extracting features based on the variable region amino acid sequence of the sample antibody to be predicted; and inputting the extracted features into an antibody immunogenicity prediction model based on a machine learning integrated model to obtain a prediction result of the antibody immunogenicity.
[0024] According to a fourth aspect of the present disclosure, a non-transitory computer-readable storage medium is provided for storing a computer program. The computer program includes instructions. When executed by a processor of an electronic device, the instructions cause the electronic device to implement the antibody immunogenicity prediction method according to the first aspect of the present disclosure or the method for constructing an antibody immunogenicity prediction model based on a machine learning integrated model according to the second aspect of the present disclosure.
[0025] According to a fifth aspect of the present disclosure, a computer system is provided. The computer system includes: a processor, a memory, and a computer program. The computer program is stored in the memory and is configured to be executed by the processor. The computer program includes instructions for implementing the antibody immunogenicity prediction method according to the first aspect of the present disclosure or the method for constructing an antibody immunogenicity prediction model based on a machine learning integrated model according to the second aspect of the present disclosure.
[0026] The immunogenicity prediction method proposed in the present disclosure is based on the amino acid sequence of the antibody variable region, and uses the embedding layer output by the ProtT5-XL-UniRef50 pre-training model as the sequence feature. The extracted features are then compressed using the Sentence-T5 Encoder-only mean method, and the extracted features are subjected to dimensionality reduction processing using principal component analysis (PCA) to obtain the final features. AutoGluon is used to integrate multiple sub-models and train the final model. The model is used to predict the immunogenicity of the antibody, and the prediction results are significantly better than the existing methods. The model constructed in the present disclosure is based on 199 approved therapeutic antibody drug samples, of which 177 samples are used as training samples for model construction and verification. The model construction and verification adopt the leave-one-out cross-validation method, and the remaining 22 samples are used as independent test sets to test the model. Using the trained model and the extracted features, the high / low immunogenicity of the antibody can be predicted, and antibody drugs predicted to be low immunogenic may be a better choice for clinical development. BRIEF DESCRIPTION OF THE DRAWINGS
[0027] The present disclosure will be more fully understood from the following detailed description taken in conjunction with the accompanying drawings, in which like elements are numbered in a similar manner, and in which:
[0028] FIG1 is a flowchart of a method for predicting immunogenicity according to an embodiment of the present disclosure.
[0029] FIG2 is a schematic diagram illustrating the association between the immunogenicity prediction method and the prediction model construction method according to an embodiment of the present disclosure.
[0030] FIG3 is a more detailed flowchart of the antibody feature extraction method according to an embodiment of the present disclosure.
[0031] FIG4 is a more detailed flowchart of the method for constructing an immunogenicity prediction model based on a machine learning ensemble model used in the present disclosure.
[0032] FIG5 is a ROC curve diagram of the classification performance using the prediction model disclosed herein.
[0033] FIG6 shows a schematic block diagram of a system for predicting antibody immunogenicity according to the present disclosure. DETAILED DESCRIPTION
[0034] Unless otherwise defined, technical and scientific terms used in the present disclosure have the same meaning as commonly understood by one of ordinary skill in the art to which the present disclosure belongs.
[0035] The technical solutions of the present disclosure are further described in detail below through examples and in conjunction with the accompanying drawings. Unless otherwise stated, the methods and materials of the embodiments described below are conventional products that can be purchased on the market. Those skilled in the art will understand that the methods and materials described below are merely exemplary and should not be construed as limiting the scope of the present disclosure.
[0036] From the overall perspective of the prediction method, FIG1 shows a flowchart of the method for predicting antibody immunogenicity according to an embodiment of the present disclosure.
[0037] As shown in FIG1 , the antibody immunogenicity prediction method 100 according to an embodiment of the present disclosure begins at step S110 , where feature extraction is performed based on the amino acid sequence of the antibody variable region.
[0038] Next, in step S120, the features extracted in step S110 are input into an antibody immunogenicity prediction model based on a machine learning integrated model to obtain a prediction result of the antibody immunogenicity.
[0039] As mentioned above, most existing computational tools for antibody immunogenicity predict B-cell and T-cell epitopes, but are unable to provide a macroscopic view of the clinically relevant immune response of antibody drugs after injection into the body. Consequently, very few tools exist for predicting the clinical immunogenicity of antibody drugs. While the prediction tool PITHA can be used to predict the high or low clinical immunogenicity of therapeutic antibody drugs, it requires the three-dimensional (3D) structure of the antibody as input to extract structural features.
[0040] The difference between the present disclosure and the above-mentioned prior art is that only the variable region amino acid sequence of the antibody is provided as the input of the prediction model, and a machine learning integrated model is adopted in the prediction model, so that the high / low immunogenicity of the antibody can be predicted quickly and accurately by calculation, providing assistance for immunogenicity screening in the early stage of therapeutic antibody drug development, greatly saving time, materials and labor costs.
[0041] Figure 2 shows a schematic diagram 200 illustrating the association between the immunogenicity prediction method and the prediction model construction method according to an embodiment of the present disclosure. In the schematic diagram 200 shown in Figure 2, the left side of the dotted line shows the model construction method, and the right side of the dotted line shows the method of using the model for prediction.
[0042] The process on the right side of the dotted line in Figure 2 is actually the antibody immunogenicity prediction method according to an embodiment of the present disclosure shown in Figure 1. The process on the left side of the dotted line in Figure 2 will be described in detail below (for example, Figure 4 and its corresponding text description). As can be seen from Figure 2, both model construction and the actual prediction process require the use of a feature extraction method based on the amino acid sequence of the antibody variable region, which will be described in detail below (for example, Figure 3 and its corresponding text description). For the training sample, it includes both the variable region amino acid sequence of the sample antibody and the immunogenicity data of the antibody. As described later, the immunogenicity data of the antibody can be highly immunogenic or low immunogenic. Therefore, during the model construction process, the sequence of the antibody in the training sample is used to extract features. By using the extracted features as input data and the antibody immunogenicity data in the training sample as output data, the machine learning integrated model (for example, using AutoGluon, one of the AutoML frameworks, to integrate multiple sub-models) is fully trained to obtain the final antibody immunogenicity prediction model based on the machine learning integrated model. For the constructed prediction model, it can be put into the actual work of antibody immunogenicity prediction, that is, the process on the right side of the dotted line in Figure 2.
[0043] The following describes in detail the detailed processes of the antibody feature extraction method based on variable region amino acid sequences and the method for constructing an antibody immunogenicity prediction model based on a machine learning integrated model according to the embodiments of the present disclosure.
[0044] FIG3 is a more detailed flowchart of an antibody feature extraction method 300 according to an embodiment of the present disclosure.
[0045] Generally, step S110 of the prediction method in Figure 1, i.e., feature extraction based on the antibody variable region amino acid sequence, can first extract feature vectors for the antibody heavy chain variable region amino acid sequence and the antibody light chain variable region amino acid sequence. More specifically, using the antibody heavy chain variable region amino acid sequence as input, a protein pre-trained model is used to extract a feature matrix; then, the extracted feature matrix is compressed into a feature vector. Similarly, using the antibody light chain variable region amino acid sequence as input, a protein sequence pre-trained model is used to extract a feature matrix; then, the extracted feature matrix is compressed into a feature vector.
[0046] The protein pre-training model described here can be implemented by a combination of one or more of the following models: ESM-1b, UniRep, ProteinBert, TAPE, ProtGPT2, ProtTXL, ProtBert, ProtXLNet, ProtAlbert, ProtElectra, ProtT5-XL-BFD, ProtT5-XL-UniRef50 and ProtT5-XXL.
[0047] Preferably, ProtT5-XL-BFD or ProtT5-XL-UniRef50 can be used as the protein pre-training model.
[0048] In a preferred embodiment of the present disclosure, the protein sequence pre-training model can be a ProtT5-XL-UniRef50 pre-training model. ProtT5-XL-UniRef50 is a protein sequence pre-training model using a masked language modeling (MLM) objective. Reference may be made to A.Elnaggar et al., “ProtTrans: Toward Understanding the Language of Life Through Self-Supervised Learning”, in IEEE Transactions on Pattern Analysis and Machine Intelligence, vol.44, no.10, pp.7112-7127, 1 Oct.2022, doi: 10.1109 / TPAMI.2021.3095381. By reference herein, the entire contents of the above-mentioned document are incorporated into the present disclosure to make it a part of the contents of the present disclosure. In addition, the source code of the ProtT5-XL-UniRef50 model and its more detailed application can be found at https: / / github.com / agemagician / ProtTrans, where the model was also first published. ProtT5-XL-UniRef50 is based on the t5-3b model and is pre-trained on a large number of protein sequences in a self-supervised manner. This model can be used for protein feature extraction and preferably uses features extracted from the encoder.
[0049] As shown in FIG3 , the antibody feature extraction method 300 according to an embodiment of the present disclosure may begin at step S310 . That is, as described above, according to a preferred embodiment of the present disclosure, in step S310 , the antibody heavy chain variable region amino acid sequence is used as input, and the ProtT5-XL-UniRef50 pre-trained model is used to extract the embedding layer vector output by the encoder. Each amino acid corresponds to a 1024-dimensional feature vector, and each sequence corresponds to an L×1024-dimensional feature matrix, where L is the length of the antibody heavy chain variable region amino acid sequence.
[0050] Although the ProtT5-XL-UniRef50 pre-training model is used to extract the feature matrix of the antibody variable region amino acid sequence in the preferred embodiment of the present disclosure, when the input and output are essentially unchanged and the functions are similar and the effects are similar or better, those skilled in the art are fully motivated to use another or newly developed protein pre-training model to replace the ProtT5-XL-UniRef50 pre-training model used in the present disclosure. Those skilled in the art should understand that even after such a pre-training model is replaced, the replaced solution still falls within the scope of protection claimed in the present disclosure.
[0051] Next, the obtained feature matrix can be compressed. Specifically, the feature matrix obtained in the above steps can be averaged by column to obtain a 1024-dimensional feature vector.
[0052] As shown in FIG3 , in a preferred embodiment of the present disclosure, in step S320 , the feature matrix obtained in step S310 is compressed, that is, the Sentence-T5 Encoder-only mean processing method can be used to average the values of each dimension of the embedding layer feature values of all amino acids output by the encoder by column to obtain a 1024-dimensional feature vector. The Sentence-T5 Encoder-only mean is one way to process the encoder output features in Sentence-T5. For more information about the Sentence-T5 Encoder-only mean, see Jianmo Ni, Gustavo Hernandez Abrego, Noah Constant, Ji Ma, Keith Hall, Daniel Cer, and Yinfei Yang, “Sentence-T5: Scalable Sentence Encoders from Pre-trained Text-to-Text Models”, in Findings of the Association for Computational Linguistics: ACL 2022, pages 1864–1874, Dublin, Ireland; Association for Computational Linguistics; doi:10.18653 / v1 / 2022.findings-acl.146. The entire contents of the above document are hereby incorporated into the present disclosure by reference, making it a part of the content of the present disclosure.
[0053] Those skilled in the art should note that although Sentence-T5 Encoder-only mean is a disclosed technology, its application in the method of the present disclosure is novel because the application field involved in the present disclosure is completely different from the application field involved in the original disclosure of Sentence-T5 Encoder-only mean. The present disclosure uses Sentence-T5 Encoder-only mean to compress the extracted feature matrix, so that the extracted feature vector more deeply integrates the characteristics of the antibody variable region amino acids, which is helpful for the subsequent construction of a better performance prediction model based on this feature, and ultimately obtains a more accurate antibody immunogenicity prediction result.
[0054] For the amino acid sequence of the antibody light chain variable region, in step S330 , feature extraction is performed in the same manner as in steps S310 and S320 to obtain a 1024-dimensional feature vector.
[0055] More specifically, the amino acid sequence of the variable region of the antibody light chain is used as input, and the ProtT5-XL-UniRef50 pre-trained model is used to extract the embedding layer vector output by the encoder. Among them, each amino acid corresponds to a 1024-dimensional feature vector, and each sequence corresponds to an L'×1024-dimensional feature matrix, where L' is the length of the amino acid sequence of the variable region of the antibody light chain. Next, the feature matrix obtained above can be averaged by column to obtain a 1024-dimensional feature vector. Specifically, the Sentence-T5 Encoder-only mean processing method can be used to average the values of each dimension of the embedding layer feature values of all amino acids output by the encoder by column to obtain a 1024-dimensional feature vector.
[0056] After extracting feature vectors for the antibody heavy chain variable region amino acid sequence and the antibody light chain variable region amino acid sequence, respectively, the feature vector extracted for the antibody heavy chain variable region amino acid sequence and the feature vector extracted for the antibody light chain variable region amino acid sequence are spliced according to the method disclosed in the present invention to obtain the feature vector of the antibody.
[0057] As shown in FIG3 , in step S340 , the two 1024-dimensional feature vectors obtained in steps S320 and S330 are concatenated to obtain a 2048-dimensional feature vector.
[0058] It is understandable that the splicing order of the feature vectors extracted for the amino acid sequence of the antibody heavy chain variable region and the feature vectors extracted for the amino acid sequence of the antibody light chain variable region is not limited in this embodiment. It is only necessary to ensure that the splicing order of the feature vectors extracted for the amino acid sequence of the antibody heavy chain variable region and the feature vectors extracted for the amino acid sequence of the antibody light chain variable region remain consistent during the model construction and actual prediction process. Illustratively, during the model construction process, the feature vectors extracted from the amino acid sequence of the heavy chain variable region of the sample antibody and the feature vectors extracted from the amino acid sequence of the light chain variable region of the sample antibody are spliced in order to obtain the feature vector of the sample antibody. Then, during the actual prediction process, the feature vectors extracted from the amino acid sequence of the heavy chain variable region of the antibody and the feature vectors extracted from the amino acid sequence of the light chain variable region of the antibody are spliced in this order to obtain the feature vector of the sample antibody to be predicted. Conversely, during the model construction process, the feature vectors extracted from the amino acid sequence of the light chain variable region of the sample antibody and the feature vectors extracted from the amino acid sequence of the heavy chain variable region of the sample antibody are spliced in this order to obtain the feature vector of the sample antibody. Then, during the actual prediction process, the feature vectors extracted from the amino acid sequence of the light chain variable region of the antibody and the feature vectors extracted from the amino acid sequence of the heavy chain variable region of the antibody are spliced in this order to obtain the feature vector of the sample antibody to be predicted.
[0059] Next, the method may further include step S350, in which the feature vector of the antibody obtained in step S340 is reduced in dimension to obtain the feature vector of the antibody after dimension reduction. For example, the principal component analysis (PCA) algorithm ( A, Ratajczak W. Principal components analysis (PCA) [J]. Computers & Geosciences, 1993, 19 (3): 303-342.) performs dimension reduction on the feature vector of the antibody to obtain the feature vector of the antibody after dimension reduction.
[0060] For example, as shown in FIG3 , in step S350 , the PCA model may be used to reduce the dimension of the 2048-dimensional feature vector obtained after concatenation in step S340 with a dimension reduction parameter of 100 to obtain a 100-dimensional feature vector.
[0061] It should be noted here that the PCA model used in the method of the present disclosure can be constructed by the following steps:
[0062] (1) Using the features of the training samples as input, the PCA model is fitted;
[0063] (2) Using the dimensions after different dimensionality reduction as parameters, the PCA model parameters are optimized and selected.
[0064] For example, the training samples can use the same training samples as those used in the prediction model construction method of the present disclosure, as described later, for example, using all 177 training samples, and after the feature extraction step as described above, using the features obtained in step S340 as input to fit the PCA model.
[0065] Although the PCA method is used to reduce the dimensionality of feature vectors in the preferred embodiment of the present disclosure, those skilled in the art are fully motivated to adopt another or newly developed dimensionality reduction model to replace the PCA model used in the present disclosure, provided that the input and output feature vectors do not change in essence and the method functions similarly and achieves similar or better results. Those skilled in the art should understand that even after such a dimensionality reduction model is replaced, the replaced solution still falls within the scope of protection claimed in the present disclosure.
[0066] Through the above series of steps S310-S350, according to the method disclosed herein, the desired features can be extracted based solely on the variable region amino acid sequence of the antibody. Thus, these features can be used as input to predict the immunogenicity of the antibody using a machine learning ensemble model, outputting a predicted antibody immunogenicity result.
[0067] On the other hand, as shown in Figure 2, the above series of steps can also be used to extract features based on the amino acid sequence of the antibody variable region of the training samples, so that the extracted features can be used as input data and input into a machine learning integration model constructed by, for example, AutoGluon. The immunogenicity data of the corresponding antibodies in the training samples are used as output data to train a prediction model based on the machine learning integration model, so as to finally determine the model parameters, thereby constructing an antibody immunogenicity prediction model with relatively satisfactory accuracy.
[0068] FIG4 is a more detailed flowchart of a method 400 for constructing an antibody immunogenicity prediction model based on a machine learning integrated model used in the present disclosure.
[0069] Those skilled in the art will appreciate that, as shown in Figure 4, in step S410, a data set comprising a plurality of training samples should first be obtained. Specifically, each of the plurality of training samples in the data set comprises the variable region amino acid sequence of the sample antibody and immunogenicity data of the sample antibody.
[0070] As mentioned above, since the data sets used in building models by existing methods, including the prediction tool PITHA, are relatively small, there is much room for improvement in the accuracy of predictions.
[0071] In a preferred embodiment of the present disclosure, data acquisition includes at least two approaches:
[0072] 1. Download the variable region sequences of approved therapeutic antibodies from the U.S. Patent and Trademark Office (patft.uspto.gov) and the KEGG drug database (https: / / www.genome.jp / KEGG / drug);
[0073] 2. Obtain immunogenicity data of therapeutic antibodies from the US FDA website (https: / / www.accessdata.fda.gov / scripts / cder / daf).
[0074] In a preferred embodiment of the present disclosure, the immunogenicity data of the antibody may be high immunogenicity or low immunogenicity.
[0075] Accordingly, for the therapeutic antibodies in the data or data sets obtained by the above methods, the immunogenicity is divided into high and low according to the proportion of patients who produce antigen-antibody response (AAR) during clinical treatment.
[0076] Specifically, in a preferred embodiment, the threshold for classifying immunogenicity as high or low is 2%. That is, if the number of patients with an AAR produced by a therapeutic antibody during clinical treatment is less than 2%, the immunogenicity data of the antibody is low, and the remaining patients are high.
[0077] In addition, in order to compare with the existing method PITHA, all 199 therapeutic antibodies obtained were divided into a training set (containing 177 training samples) and an independent test set (containing 22 test samples).
[0078] Then, as mentioned above, feature extraction is required for the variable region amino acid sequence of each training sample to generate input data for training the prediction model. As shown in step S420 in Figure 4, feature extraction is performed on all training samples according to the method of Figure 3.
[0079] Next, the extracted features are used as input data, and the immunogenicity data of the corresponding antibodies in the training samples are used as the output data of the machine learning integration model to train the machine learning integration model to obtain the final antibody immunogenicity prediction model based on the machine learning integration model.
[0080] Specifically, automated machine learning (AutoML) can be used to build and train a machine learning ensemble model based on one or more machine learning models. For example, AutoGluon is an AutoML framework. Other machine learning methods are also feasible.
[0081] Those skilled in the art will appreciate that implementing a machine learning model involves significant complexity and learning. It also requires extensive domain expertise to generate and compare multiple models before selecting the optimal model to ultimately achieve the optimal model. Automated machine learning can simplify this process. AutoML is used to automate the process of building the entire machine learning pipeline with minimal human intervention. Furthermore, using AutoML significantly speeds up training on small and medium-sized datasets compared to larger datasets.
[0082] AutoGluon is a popular AutoML framework. This open-source AutoML toolkit, developed by AWS, helps achieve strong predictive performance across a variety of machine learning and deep learning models.
[0083] As shown in Figure 4, in step S430, the AutoGluon tool is used to build and train a machine learning integration model based on one or more models of NeuralNetFastAI, LightGBMLarge, CatBoost, LightGBM, XGBoost, LightGBMXT, NeuralNetTorch, KNeighborsUnif, KNeighborsDist, ExtraTreesEntr, ExtraTreesGini, RandomForestGini, and RandomForestEntr models. In a specific embodiment, in step S430, the AutoGluon tool is used to build and train a machine learning integration model based on XGBoost and NeuralNetFastAI models.
[0084] As previously mentioned, during the model training process, the extracted features are used as the input data of the machine learning ensemble model, and the antibody immunogenicity data (e.g., high vs. low) is used as the output data of the machine learning ensemble model to train the model. As the model parameters are continuously optimized, the output data of the machine learning ensemble model will be consistent with the immunogenicity data of the sample antibodies.
[0085] According to a preferred embodiment of the present disclosure, a leave-one-out method is used to perform cross-validation and determine the final model parameters.
[0086] As mentioned below, the final model can be evaluated using an independent test set and constructed to predict immunogenicity.
[0087] Those skilled in the art should understand that the prediction model obtained by, for example, the method of FIG. 4 can be used in the prediction method shown in FIG. 1 .
[0088] This paper uses 199 FDA-approved therapeutic antibodies and their immunogenicity data during clinical use to build a model for validation and testing. The accuracy (ACC) obtained by using leave-one-out cross-validation on the training set is 0.8588.
[0089] Figure 5 is a receiver operating characteristic (ROC) curve for the classification performance of the prediction model using the present disclosure. The horizontal axis represents the false positive rate, the vertical axis represents the true positive rate, and the area under the curve corresponds to the AUROC (Area Under ROC) value. A larger value represents better classifier performance, with a maximum value of 1. As shown in Figure 5, the AUROC value of the classification performance indicator of the method established by the present disclosure is 0.8021, indicating that the method has good classification performance.
[0090] In addition, the prediction model constructed in the present disclosure was tested on an independent test set and compared with the existing method PITHA (address: http: / / sysbio.unl.edu / PITHA / ), and the results are shown in Table 1.
[0091] Table 1 Comparison of the disclosed model with existing methods
[0092] The formulas for each indicator in the table are listed as follows:
[0093] Accuracy (ACC) refers to the ratio of correctly classified samples to the total number of samples. The accuracy calculation formula is: ACC = (TP + TN) / (P + N). Generally, the higher the accuracy, the better the classifier performance.
[0094] Recall refers to the ratio of the number of correctly predicted positive samples to the total number of positive samples. The recall calculation formula is: Recall = TP / (TP + FN), and the value range is between 0 and 1. The higher the recall, the stronger the model's ability to find correct positive samples.
[0095] Precision refers to the proportion of correctly predicted results among all the results predicted as positive samples. The precision calculation formula is: Precision = TP / (TP + FP). The closer the value is to 1, the better the performance.
[0096] The F1 score is defined as the harmonic mean of precision and recall. The calculation formula for the F1 score is: F1 = 2 / ((1 / Precision) + (1 / Recall)). The value range is between 0 and 1. A larger F1 score indicates a higher model quality.
[0097] MCC (Matthews correlation coefficient) is an indicator commonly used to evaluate the predictive performance of classification models. It describes the correlation coefficient between the actual classification and the predicted classification. The calculation formula of MCC is: MCC = (TP*TN-FP*FN) / (sqrt((TP+FP)*(TP+FN)*(TN+FP)*(TN+FN))). The value range of MCC is between -1 and +1, where +1 indicates a perfect prediction, 0 indicates a random prediction, and -1 indicates that the prediction is completely inconsistent with the actual observation.
[0098] Among them, TP (True Positive) and TN (True Negative) represent the number of correctly classified positive samples and negative samples, FP (False Positive) and FN (False Negative) represent the number of incorrectly classified positive samples and negative samples, and P and N represent the number of positive samples and negative samples.
[0099] From Table 1, it can be seen that the method disclosed in the present invention performs well in various evaluation indicators and is significantly better than the existing methods.
[0100] Those skilled in the art will appreciate that, based on the antibody immunogenicity prediction method disclosed herein, a system for predicting antibody immunogenicity can be developed. FIG6 shows a schematic block diagram of a system for predicting antibody immunogenicity according to the present disclosure. Specifically, the system 600 for predicting antibody immunogenicity includes an acquisition module 610 and a processing module 620.
[0101] The acquisition module 610 is used to acquire a sample to be predicted, where the sample to be predicted includes the variable region amino acid sequence of the antibody to be predicted.
[0102] Processing module 620 is configured to obtain a prediction result for the immunogenicity of the sample antibody to be predicted by inputting the variable region amino acid sequence of the sample antibody to be predicted. As shown in FIG6 , processing module 620 may further include: a feature extraction submodule 621 for extracting features based on the variable region amino acid sequence of the sample antibody to be predicted; and a model prediction submodule 622 for inputting the extracted features into an antibody immunogenicity prediction model based on a machine learning ensemble model to obtain a prediction result for the antibody immunogenicity.
[0103] Those skilled in the art will appreciate that the system for predicting antibody immunogenicity described above can be implemented by a computer. For example, the following operations can be implemented by executing a computer program: obtaining a sample to be predicted, wherein the sample to be predicted includes the variable region amino acid sequence of the sample antibody to be predicted; obtaining a prediction result of the immunogenicity of the sample antibody to be predicted by inputting the variable region amino acid sequence of the sample antibody to be predicted, wherein the operation further includes the following sub-operations: extracting features based on the variable region amino acid sequence of the sample antibody to be predicted; inputting the extracted features into an antibody immunogenicity prediction model based on a machine learning integrated model to obtain a prediction result of the antibody immunogenicity. That is, although in the system for predicting antibody immunogenicity according to the present disclosure, the various operational functions corresponding to the prediction method are described as modules or submodules, those skilled in the art will appreciate that such modules or submodules may not be composed of circuit components or other physical components, but rather functional modules constructed by computer programs.
[0104] In addition, those skilled in the art will recognize that the methods disclosed herein can be implemented as computer programs. As described above in conjunction with the accompanying drawings, the methods of the above embodiments are executed by one or more programs, and the instructions in the programs cause a computer or processor to execute the algorithms described in conjunction with the accompanying drawings. These programs can be stored and provided to a computer or processor using various types of non-transitory computer-readable media. Non-transitory computer-readable media include various types of tangible storage media. Examples of non-transitory computer-readable media include magnetic recording media (such as floppy disks, magnetic tapes, and hard disk drives), magneto-optical recording media (such as magneto-optical disks), CD-ROMs (compact disk read-only memories), CD-Rs, CD-R / Ws, and semiconductor memories (such as ROMs, PROMs (programmable ROMs), EPROMs (erasable PROMs), flash ROMs, and RAMs (random access memories)). Furthermore, these programs can be provided to a computer using various types of transient computer-readable media. Examples of transient computer-readable media include electrical signals, optical signals, and electromagnetic waves. Transient computer-readable media can be used to provide programs to a computer via wired communication paths such as wires and optical fibers or wireless communication paths.
[0105] For example, according to one embodiment of the present disclosure, a non-transitory computer-readable storage medium may be provided for storing a computer program, wherein the computer program includes instructions that, when executed by a processor of an electronic device, enable the electronic device to implement the antibody immunogenicity prediction method as described above.
[0106] In addition, according to the present disclosure, a computer system may be provided, comprising: a processor; a memory; and a computer program. The computer program is stored in the memory and configured to be executed by the processor. The computer program includes instructions for implementing the above-described method for predicting antibody immunogenicity.
[0107] On the other hand, according to one embodiment of the present disclosure, a non-transitory computer-readable storage medium may be provided for storing a computer program, wherein the computer program includes instructions that, when executed by a processor of an electronic device, enable the electronic device to implement the method for constructing an antibody immunogenicity prediction model based on a machine learning integrated model as described above.
[0108] In addition, according to the present disclosure, a computer system may also be provided, comprising: a processor; a memory; and a computer program. The computer program is stored in the memory and configured to be executed by the processor. The computer program includes instructions for implementing the method for constructing an antibody immunogenicity prediction model based on the machine learning ensemble model described above.
[0109] The implementation methods of the present disclosure are not limited to those described in the above embodiments. Without departing from the spirit and scope of the present disclosure, ordinary technicians in this field can make various changes and improvements to the present disclosure in form and details, and these are all considered to fall within the scope of protection of the present disclosure.
Claims
1. A method for predicting antibody immunogenicity, characterized in that: The method comprises: Feature extraction based on the amino acid sequence of the antibody variable region; The extracted features are input into the antibody immunogenicity prediction model based on the machine learning ensemble model to obtain the prediction results of antibody immunogenicity.
2. The method according to claim 1, characterized in that The feature extraction based on the antibody variable region amino acid sequence includes: Extract feature vectors for the amino acid sequence of the variable region of the antibody heavy chain; Extract feature vectors based on the amino acid sequence of the variable region of the antibody light chain; The feature vector extracted from the amino acid sequence of the antibody heavy chain variable region and the feature vector extracted from the amino acid sequence of the antibody light chain variable region are spliced to obtain the feature vector of the antibody.
3. The method according to claim 2, characterized in that The method of extracting a feature vector for the amino acid sequence of the variable region of the antibody heavy chain comprises: taking the amino acid sequence of the variable region of the antibody heavy chain as input, extracting a feature matrix using a protein pre-training model, and compressing the extracted feature matrix into a feature vector; The method of extracting a feature vector for an antibody light chain variable region amino acid sequence includes: taking an antibody light chain variable region amino acid sequence as input, extracting a feature matrix using a protein pre-training model, and compressing the extracted feature matrix into a feature vector.
4. The method according to claim 3, characterized in that: The protein pre-training model is implemented by a combination of one or more of the following models: ESM-1b, UniRep, ProteinBert, TAPE, ProtGPT2, ProtTXL, ProtBert, ProtXLNet, ProtAlbert, ProtElectra, ProtT5-XL-BFD, ProtT5-XL-UniRef50 and ProtT5-XXL; preferably, the protein pre-training model is ProtT5-XL-BFD or ProtT5-XL-UniRef50; more preferably, the protein pre-training model is ProtT5-XL-UniRef50.
5. The method according to claim 4, characterized in that The step of extracting a feature vector from the amino acid sequence of the antibody heavy chain variable region comprises: The amino acid sequence of the variable region of the antibody heavy chain was used as input, and the ProtT5-XL-UniRef50 pre-trained model was used to extract the embedding layer vector output by the encoder, where each amino acid corresponds to a 1024-dimensional vector and each sequence corresponds to an L×1024-dimensional feature matrix, where L is the length of the amino acid sequence of the variable region of the antibody heavy chain. Take the average value of the above feature matrix by column to obtain a 1024-dimensional feature vector; The method of extracting a feature vector for the amino acid sequence of the antibody light chain variable region comprises: The amino acid sequence of the variable region of the antibody light chain was used as input, and the ProtT5-XL-UniRef50 pre-trained model was used to extract the embedding layer vector output by the encoder, where each amino acid corresponds to a 1024-dimensional vector and each sequence corresponds to an L'×1024-dimensional feature matrix, where L' is the length of the amino acid sequence of the variable region of the antibody light chain. Take the average value of the above feature matrix by column to obtain a 1024-dimensional feature vector.
6. The method according to claim 5, characterized in that The step of taking the average value of the feature matrix by columns to obtain a 1024-dimensional feature vector further includes: The Sentence-T5Encoder-onlymean method is used to average the values of each dimension of the embedding layer feature values of all amino acids output by the encoder by column to obtain a 1024-dimensional feature vector.
7. The method according to claim 5, characterized in that The feature vector extracted from the amino acid sequence of the antibody heavy chain variable region and the feature vector extracted from the amino acid sequence of the antibody light chain variable region are spliced to obtain the feature vector of the antibody, comprising: The 1024-dimensional feature vector extracted from the amino acid sequence of the antibody heavy chain variable region and the 1024-dimensional feature vector extracted from the amino acid sequence of the antibody light chain variable region are concatenated to obtain a 2048-dimensional feature vector.
8. The method according to claim 2, characterized in that: The method further comprises: The principal component analysis method is used to perform dimension reduction processing on the characteristic vector of the antibody to obtain the characteristic vector of the antibody after dimension reduction.
9. The method according to claim 1, characterized in that: The antibody immunogenicity prediction model based on the machine learning integrated model is constructed through the following steps: Acquire a data set, the data set comprising a plurality of training samples, each of the training samples comprising a variable region amino acid sequence and immunogenicity data of a sample antibody; Extracting features from the variable region amino acid sequence of each of the sample antibodies; The extracted features are used as input data, and the antibody immunogenicity data is used as output data of the machine learning integrated model to train the machine learning integrated model to obtain the final antibody immunogenicity prediction model based on the machine learning integrated model.
10. The method according to claim 9, characterized in that The training of the machine learning integrated model includes: using an automatic machine learning method to build and train the machine learning integrated model based on one or more machine learning models.
11. The method according to claim 10, characterized in that The training of the machine learning integration model includes: using the AutoGluon tool to build and train the machine learning integration model based on one or more models of NeuralNetFastAI, LightGBMLarge, CatBoost, LightGBM, XGBoost, LightGBMXT, NeuralNetTorch, KNeighborsUnif, KNeighborsDist, ExtraTreesEntr, ExtraTreesGini, RandomForestGini and RandomForestEntr models.
12. A method for constructing an antibody immunogenicity prediction model based on a machine learning integrated model, characterized in that: The construction method comprises: Acquire a data set, the data set comprising a plurality of training samples, each of the training samples comprising a variable region amino acid sequence and immunogenicity data of a sample antibody; Extracting features from the variable region amino acid sequence of each of the sample antibodies; The extracted features are used as input data, and the antibody immunogenicity data is used as output data of the machine learning integrated model to train the machine learning integrated model to obtain the final antibody immunogenicity prediction model based on the machine learning integrated model.
13. The method according to claim 12, characterized in that The feature extraction of the variable region amino acid sequence of each sample antibody comprises: Extract feature vectors based on the amino acid sequence of the heavy chain variable region of the sample antibody; Extract feature vectors based on the amino acid sequence of the light chain variable region of the sample antibody; The feature vector extracted from the amino acid sequence of the heavy chain variable region of the sample antibody and the feature vector extracted from the amino acid sequence of the light chain variable region of the sample antibody are spliced to obtain the feature vector of the sample antibody.
14. The method according to claim 13, characterized in that The extracting of feature vectors for the amino acid sequence of the heavy chain variable region of the sample antibody comprises: taking the amino acid sequence of the heavy chain variable region of the sample antibody as input, extracting a feature matrix using a protein pre-training model, and compressing the extracted feature matrix into a feature vector; The method of extracting a feature vector for the amino acid sequence of the light chain variable region of a sample antibody includes: taking the amino acid sequence of the light chain variable region of the antibody as input, extracting a feature matrix using a protein pre-training model, and compressing the extracted feature matrix into a feature vector.
15. The method according to claim 14, characterized in that The protein pre-training model is implemented by a combination of one or more of the following models: ESM-1b, UniRep, ProteinBert, TAPE, ProtGPT2, ProtTXL, ProtBert, ProtXLNet, ProtAlbert, ProtElectra, ProtT5-XL-BFD, ProtT5-XL-UniRef50 and ProtT5-XXL; preferably, the protein pre-training model is ProtT5-XL-BFD or ProtT5-XL-UniRef50; more preferably, the protein pre-training model is ProtT5-XL-UniRef50.
16. The method according to claim 15, characterized in that The step of extracting a feature vector from the amino acid sequence of the heavy chain variable region of the sample antibody comprises: The amino acid sequence of the variable region of the heavy chain of the sample antibody is used as input, and the ProtT5-XL-UniRef50 pre-trained model is used to extract the embedding layer vector output by the encoder, where each amino acid corresponds to a 1024-dimensional vector and each sequence corresponds to an L×1024-dimensional feature matrix, where L is the length of the amino acid sequence of the variable region of the heavy chain of the sample antibody; Take the average value of the above feature matrix by column to get a 1024-dimensional feature vector. The method of extracting a feature vector from the amino acid sequence of the light chain variable region of the sample antibody comprises: The amino acid sequence of the light chain variable region of the sample antibody is used as input, and the ProtT5-XL-UniRef50 pre-trained model is used to extract the embedding layer vector output by the encoder, where each amino acid corresponds to a 1024-dimensional vector and each sequence corresponds to an L'×1024-dimensional feature matrix, where L' is the length of the amino acid sequence of the light chain variable region of the sample antibody; Take the average value of the above feature matrix by column to obtain a 1024-dimensional feature vector.
17. The method according to claim 16, characterized in that The step of taking the average value of the feature matrix by columns to obtain a 1024-dimensional feature vector further includes: The Sentence-T5Encoder-onlymean method is used to average the values of each dimension of the embedding layer feature values of all amino acids output by the encoder by column to obtain a 1024-dimensional feature vector.
18. The method according to claim 16, characterized in that The feature vector extracted from the amino acid sequence of the heavy chain variable region of the sample antibody and the feature vector extracted from the amino acid sequence of the light chain variable region of the sample antibody are spliced to obtain the feature vector of the antibody, comprising: The 1024-dimensional feature vector extracted from the amino acid sequence of the heavy chain variable region of the sample antibody and the 1024-dimensional feature vector extracted from the amino acid sequence of the light chain variable region of the sample antibody are concatenated to obtain a 2048-dimensional feature vector.
19. The method according to claim 13, characterized in that The method further comprises: The characteristic vector of the antibody is subjected to dimension reduction processing using a principal component analysis method to obtain the characteristic vector of the sample antibody after dimension reduction.
20. The method according to claim 12, characterized in that The training of the machine learning integrated model includes: using an automatic machine learning method to build and train the machine learning integrated model based on one or more machine learning models.
21. The method according to claim 20, characterized in that The training of the machine learning integration model includes: using the AutoGluon tool to build and train the machine learning integration model based on one or more models of NeuralNetFastAI, LightGBMLarge, CatBoost, LightGBM, XGBoost, LightGBMXT, NeuralNetTorch, KNeighborsUnif, KNeighborsDist, ExtraTreesEntr, ExtraTreesGini, RandomForestGini and RandomForestEntr models.
22. A computer-implemented system for predicting antibody immunogenicity, wherein the computer-implemented system implements the following operations by executing a computer program: Obtaining a sample to be predicted, wherein the sample to be predicted includes a variable region amino acid sequence of an antibody of the sample to be predicted; By inputting the variable region amino acid sequence of the sample antibody to be predicted, a prediction result of the immunogenicity of the sample antibody to be predicted is obtained, and the operation further includes the following sub-operations: Feature extraction is performed based on the variable region amino acid sequence of the sample antibody to be predicted; The extracted features are input into the antibody immunogenicity prediction model based on the machine learning ensemble model to obtain the prediction results of antibody immunogenicity.
23. A non-transitory computer-readable storage medium for storing a computer program, wherein the computer program comprises instructions, which, when executed by a processor of an electronic device, causes the electronic device to implement the antibody immunogenicity prediction method as described in any one of claims 1 to 11 or the method for constructing an antibody immunogenicity prediction model based on a machine learning integrated model as described in any one of claims 12 to 21.
24. A computer system, comprising: processor; Memory; and A computer program, wherein the computer program is stored in the memory and is configured to be executed by the processor, the computer program comprising instructions for implementing the antibody immunogenicity prediction method as described in any one of claims 1 to 11 or the method for constructing an antibody immunogenicity prediction model based on a machine learning integrated model as described in any one of claims 12 to 21.
Citation Information
Patent Citations
Polypeptide TCR immunogenicity prediction method based on Bi-LSTM and Self-Attention
CN112071361A
Antigen immunogenicity prediction method and system based on multi-mode depth coding
CN113762416A
Protein immunogenicity classifier construction method, prediction method, device and medium
CN115064217A
Pipe column extrusion type rubber current collector parameter prediction and optimization system
CN115238412A
Methods, models and systems related to antibody immunogenicity
WO2021147642A1
Cited By
Method and system for predicting affinity of antibody mutant and antigen
CN119446289A
Combination prediction method of complementarity determining region 3 and immune epitope
CN120319302A
